The week-two scorecard should decide whether AgriVoice earned another phase: farmers must return, understand the spoken answers, and receive human follow-up within one working day when the service escalates a question. A first voice note alone proves curiosity; repeat use, comprehension, safe escalation and measured response time show whether a cocoa-sector partner has a service worth continuing.
The scorecard begins where novelty ends
Consider Kojo, an illustrative composite, standing beside a cocoa plot late in the afternoon with a pod he has wrapped in a plastic bag. He has already sent one voice note earlier in the week and received an answer he understood. Now the markings on a second pod look different, and he is deciding whether to wait, remove affected pods, or ask someone to inspect the tree.
His first message made the service feel possible. His second message is the test.
If AgriVoice gives a reviewed answer that fits the question, the farmer can act with more confidence. If the question is uncertain, code-switched, incomplete, or touches pesticide details that have not been verified, it must stay open for a named extension officer. The bad outcome is clear: Kojo could act on incomplete advice, or wait too long while a problem spreads through the plot.
That is why a two-week pilot cannot be judged by the number of first messages alone. A farmer who returns in week two has made room for the service in a real decision. A farmer who does not return may have solved the problem elsewhere, forgotten the number, found the answer confusing, or decided the response did not help enough to use again. The scorecard needs to distinguish those possibilities before anyone calls the pilot a success.
Three measures reveal whether the service helped
The first measure is successful answer rate. AgriVoice should answer a question with approved content or escalate it correctly when it cannot answer safely. The pilot’s continue signal is at least 70% answered or correctly escalated. A confident wrong answer outweighs a high count of completed conversations.
The second is comprehension. After receiving a response, farmers should be asked plainly whether they understood it. The target is at least 80% reporting that they understood the answer. This matters because the current Twi voice was trained from scripture recordings and can sound formal. A response may arrive quickly and still fail if the farmer cannot comfortably follow it.
The third is repeat use. At least 30% of activated farmers should ask another question in week two. That threshold does not claim that every farmer will need advice every week. It asks a narrower question: when another cocoa decision appears, does the farmer return to this channel?
Kojo’s second message belongs here. If he heard the first response clearly, trusted the boundary between approved advice and escalation, and knew what would happen next, he has a reason to use the service again. If he hears a formal-sounding answer with unfamiliar phrasing, he may return to the person he already knows instead. That response is useful evidence, even when it is disappointing.
An escalation only counts when a person owns it
A safe refusal is only the beginning. The partner must name the extension officer who receives escalations and make the path visible to the farmer. The pilot scorecard sets a median response target of less than one working day.
For Kojo, the important moment comes after the system recognises that the question needs a human. His pod remains in the bag. The field decision remains unresolved. A message saying that someone will follow up means little if the queue has no owner or the officer cannot see what was asked.
The partner should review each escalation with four practical details: when it entered the queue, whether the farmer could be contacted, when the extension officer responded, and whether the response closed the question or required another visit. This gives the team a way to find the actual constraint. It may be voice recognition, reviewed content, a missing location detail, staff availability, or a handoff that farmers do not understand.
What Happens When a Cocoa Question Needs a Human Answer? explores why that ownership matters before a farming decision turns urgent.
Use the evidence to make one bounded decision
At the end of week two, the cocoa-sector partner should open the scorecard with the pilot team and ask what the evidence supports. Continue only if farmers are getting safe answers or correct escalations, understanding the spoken responses, returning in meaningful numbers, and receiving accountable follow-up. Check median end-to-end response time as well, with a target below eight seconds excluding human escalation, and measure the cost per completed question by component.
A missed threshold calls for a specific explanation, not a broad promise to improve everything. If comprehension falls short, review voice phrasing and listening feedback. If repeat use is low, inspect whether the first answers solved the right problems and whether farmers knew they could send another voice note. If escalations stall, fix staffing responsibility before changing the model. If any unsafe pesticide instruction reaches a farmer, stop and investigate before further exposure.
Kojo’s second message should end in a concrete record: an approved answer he understood, or a human response with the next safe step. That record, repeated across the small pilot cohort, is what turns a promising voice demo into evidence a partner can act on.
Comments
No comments yet.